Tesla's Vision-Only Gamble: Removing Radar Was a Risk. Was It Worth It?
In 2021, Tesla removed radar sensors. Then ultrasonics. The crash data since tells a story.
Starting in May 2021, Tesla began shipping vehicles without radar sensors. By 2022, ultrasonic sensors were gone too. The company bet everything on "Tesla Vision" — a camera-only perception system powered by neural networks. It was the most controversial hardware decision in AV history. Three years later, the data is in.
The thesis
Tesla argued that humans drive with only two eyes (cameras), so a car should be able to do the same. Critics argued that humans also have depth perception, experience, and common sense — none of which a camera has.
What Every Competitor Uses Instead
Waymo uses cameras + lidar + radar. So does Cruise (when it was operating). Zoox uses cameras + lidar + radar. Aurora uses cameras + lidar + radar. Mobileye uses cameras + radar. Tesla is the only company attempting high-level autonomous driving with cameras alone.
The Numbers Since 2021
Tesla's incident count has continued to climb. With 3,092 total incidents and 56 fatalities, and NHTSA specifically investigating camera visibility failures in its PE25012 probe (covering 3.2 million vehicles), the vision-only approach is under unprecedented scrutiny.
The investigation specifically cites failures in glare, fog, low sun angles, and dust — conditions where radar excels and cameras struggle. These aren't edge cases; they're common driving conditions that every vehicle encounters.
The Phantom Braking Connection
Phantom braking complaints surged after radar removal. Without radar to independently verify whether an object exists, cameras interpreting shadows, overpasses, and road texture changes as obstacles became a systemic problem. Radar provides a "reality check" that pure vision lacks.
The Comparison That Matters
Waymo, using its multi-sensor stack with lidar + radar + cameras, has 1,729 incidents and only 2 fatalities. Tesla, using cameras only, has 3,092 incidents and 56 fatalities. While the operating environments differ (urban robotaxi vs. consumer highway), the sensor approach appears to be a factor in the fatality differential.
Was It Worth It?
Tesla saved perhaps $200-500 per vehicle by removing radar and ultrasonic sensors. Across millions of vehicles, that's billions in savings. But 56 people are dead, NHTSA is investigating 3.2 million vehicles, and the technology still hasn't delivered on its Level 5 promise. The financial bet paid off. The safety bet is still being decided — in investigations, courtrooms, and the data.
The bottom line
Every other AV company uses sensor redundancy. Tesla chose to rely on cameras alone. The NHTSA investigation into camera failures will likely determine whether that decision was innovative — or reckless.
Related Explainer Articles
ADS vs. ADAS: The Distinction That Matters Most
2,609 fully autonomous incidents. 3,606 driver-assist incidents. Understanding the difference could save your life.
ExplainerNHTSA's Standing General Order: The Rule That Made This Database Possible
In 2021, NHTSA required all AV/ADAS manufacturers to report crashes. Here's how the SGO works and what it captures.